abstract: constellation applications and perspectives ......support and constellation platform...

1
Abstract: Constellation Constellation is a platform for different aerospace related projects that need intensive computational power. The platform supports the efforts of participating projects by providing Distributed Computation capability using BOINC[1]. Constellation will send work-units of attached projects to volunteering, idle PCs where the units are progressed. The combined power of all volunteering users will help to solve important scientific tasks in fields from astronomy to aerospace-engineering beginning from student up to university projects. The bottom line is to benefit from the generosity of the volunteers and to benefit from the accumulation of different projects, like sharing programming knowledge in distributed computing and influencing the others' simulation by its own solutions. The platform is an open space for anyone, who is an air and space enthusiast and wants to spend idle computing time or even skill for a sub-project on platform. Applications for sub-project are welcome! Project Motivation In particular in aerospace sciences simulation and optimization are major tasks to minimize structure weight, to maximize thrust or to determine system reliability. These numerical tasks need a high level of resources and hardware- support and Constellation platform offers distributed computing power to projects from professionals to students to solve their problems in an adequate time without the need to maintain their own super-computer or cluster system. In this way it is possible to get access to the needed computing power by a wider range of researchers who wouldn't have been able to create those resources because of financial, administration and operation or because of bureaucracy reasons. Constellation's goal is to expedite fundamental and applicable research and bonding researchers with citizen scientists and public. Project Implementation As a DGLR (german aerospace society) [2] academic student team at the University of Stuttgart, Germany, we set up a distributed super-computer that offers computing power by idle single desktop computers of volunteers connected via the internet. This accumulated computing power is provided to aerospace research purposes, that universities and private research institutions and projects normally can't afford. With this approach we demonstrate the capability of synergetic co-operation of scientists and scientific enthused ordinary persons. In spite of classic super-computers, that have high acquisition-, maintenance- and operation-costs and aren't state of the art after just a few years, our Constellation computer is a worldwide distributed, continuously and dynamically evolving system with high heterogeneity to which anyone can participate in who runs a computer with Window, Linux or OSX operating system. This high-performance is used to solve aerospace related tasks, such as trajectory optimization (application: TrackJack). Constellation is in its late closed test-phase and will go public in the near future. Even during test-phase 550 participating machines generated approx. 120*10^9 FLoating point OPerations per Second (120 GFLOPS). When went public it can be assumed that the amount of participating machines will increase up to 5000 with about 10 TFLOPS in one Year (reference data from RNA-World, another Rechenkraft.net e.V. project [3]). The Constellation computer represents a remarkable performance that is by far superior to high-performance clusters and it increases daily. In this way hundreds of trajectories can be simulated and optimized depending in the task's complexity per day. Participating users can follow their contribution in a clearly arranged webinterface. Finished results automatically will be send back and can be archived by the sub-projects scientists. Operation and maintenance of this system requires expenditure of time during user support, who post a variety of questions on our online-forum, so that besides the scientific aspect the project has a very interesting education component („science & society“). Applications and Perspectives There are three applications as sub-projects on the Constellation platform. o TrackJack is an ascend trajectory simulation and optimizer for space- launcher systems and space-crafts. TrackJack's origin is part of a diploma thesis in Aerospace Engineering at the University of Applied Sciences Bremen. TrackJack app is processing test-workunits and will optimize the maximum altitude of HyEnD hybrid-engine sounding rocket [4]. (Figure 2&3) o Extreme Machine's idea is not to create a huge, multimillion ton exploration rover machine, rather an extremely optimized device, under the realms of very classical mechanics, to perform specific tasks. Extreme Machine simulates the rover dynamics. (Figure 4) o On The Moon aims to simulate the process occurring at moon near the surface. It is a large scale simulation aiming to create a total model of the moon system, and to compare it with data collected from the Google Lunar X-PRIZE (GLXP)[5] mission. Abstract: VOLPEX - Parallel Execution in Volunteer environment V olpex[9-11] has the goal of enabling robust execution of communicating parallel applications on volunteer PCs. The project developed the fundamental concept of autonomous redundant processes: a logical process may have multiple distributed instances that are unaware of each other, process instances can be replicated, checkpointed or re- created independently, and individual process instances can fail without application failure. The overall application progress is dictated by the state of the fastest active instance of each process. A single replication framework can potentially provide protection from process/node and network failures, varying processing speeds, and Byzantine failures caused by software and hardware errors and malicious hosts. Two prototype programming frameworks were successfully developed and validated in this project: o Volpex Dataspace API: An abstract ``dataspace’’ with asynchronous anonymous Put/Get communication operations from autonomous redundant processes. o Volpex MPI: A highly failure resistant MPI implementation based on sender based logging. A Volpex prototype has been validated and evaluated with NAS and other benchmarks, a Map-Reduce framework, and a Replica Exchange Molecular Dynamics code, and interfaces with BOINC. Results Even though a lot of aerospace projects sounds like science-fiction we lay emphasis on applicable and fundamental work that will influence current research. So the participants will directly be involved and success in research can be feed back to the community to gratify them that leads to a good atmosphere and long-time participation. Constellation as a platform will leave it to the sub-projects to openly publish parts or the complete results. References [1] BOINC – Berkeley Open Interface for Network Computing - http://boinc.berkeley.edu [2] DGLR – Deutsche Gesellschaft für Luft- und Raumfahrt - http://stuttgart.dglr.de [3] RNA World – http://rnaworld.de/rnaworld | Rechenkraft.net e.V. - http.//.rechenkraft.net [4] HyEnD – Hybrid Engine Development - http://www.hybrid-engine-development.de [5] Google Lunar X-Prize - http://glxp.org [6] MAGE – The Marburg Ad-hoc Grid Environment - http://mage.uni-marburg.de [7] OpenFOAM – Open Field Operation and Manipulation - http://www.openfoam.com [8] Ansys – CFX, Fluent - http://ansys.com [9] VOLPEX - Parallel Execution in Volunteer environment - http://volpex.cs.uh.edu/VCP [10] 'A Robust Communication Framework for Parallel Execution on Volunteer PC Grids', by Eshwar Rohit, Hien Nguyen, Nagarajan Kanna, Jaspal Subhlok, Edgar Gabriel, Qian Wang, Margaret S. Cheung, and David Anderson, The 11th IEEE/ACM International Symposium on Cluster, Cloud, and Grid Computing , Newport Beach, CA, May 2011. [11] VolpexMPI: an MPI Library for Execution of Parallel Applications on Volatile Nodes, by T. LeBlanc, R. Anand, E. Gabriel, and J. Subhlok. In Proc. The 16th EuroPVM/MPI 2009 Conference, Espoo, Finland, 2009. Published as Springer-Verlag LNCS,5759. We want to use parallel execution for CFD analysis.We want to use OpenFOAM[7] because it's used in the scientific and academic community, it offers a wide range of solvers and the main reason is that its GNU General Public License allows the use in a distributed environment. In comparison Ansys' powerful CFX and FLUENT[8] programmes use a proprietary software licence that makes it difficult to use outside the given licence agreement. For Constellation a CFD application is intended to be used in aerodynamics and stability simulation, in engine combustion analysis for micro- and jet engines, in ramjet, scramjet and pulse detonation engine and rocket engines and motors and many more. We want to open up this important field of parallel execution for distributed computing. Currently the BOINC infrastructure Constellation uses supports independent workunits. The above applications respect this limit by using splitted tasks as workunits that doesn't rely on other workunits' result. But there are aerospace tasks, like in computational fluid dynamics (CFD), that are only feasible when the complete tasks is solved in parallel to finish in a decent time. Therefore Constellation is working to extend the system to combine the advantages of distributed computing system with that of parallel clusters to create a virtual cluster where nodes are connected via the internet. We work together with Volpex Group at University of Houston to bring parallel execution to Constellation and to other BOINC-projects, and we examine MAGE[6] of University of Marburg as an additional candidate. Andreas Hornig(1,2), Lars Bausch, Sayandeep Khan, Maximilian Palm, Uwe Beckert, Sebastian de Cillia, Matthias Fritzsche, S. Müller 1. Constellation, König-Karl-Str. 27, 70372 Stuttgart, Germany 2. corresponding author: [email protected] http://aerospaceresearch.net Figure 2 Figure 3 Figure 4 Figure 5: CFD analysis with FLUENT Figure 6: possible splitted mesh processing via internet-connected client-nodes Figure 7: VOLPEX Figure 1: Constellation website "Space is big. Really big. You just won't believe how vastly, hugely, mind- bogglingly big it is. I mean, you may think it's a long way down the road to the chemist's, but that's just peanuts to space." - The Hitchhiker's Guide to the Galaxy by Douglas Adams

Upload: others

Post on 01-Mar-2021

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Abstract: Constellation Applications and Perspectives ......support and Constellation platform offers distributed computing power to projects from professionals to students to solve

Abstract: ConstellationConstellation is a platform for different aerospace related projects that need intensive computational power.The platform supports the efforts of participating projects by providing Distributed Computation capability using BOINC[1].Constellation will send work-units of attached projects to volunteering, idle PCs where the units are progressed. The combined power of all volunteering users will help to solve important scientific tasks in fields from astronomy to aerospace-engineering beginning from student up to university projects.The bottom line is to benefit from the generosity of the volunteers and to benefit from the accumulation of different projects, like sharing programming knowledge in distributed computing and influencing the others' simulation by its own solutions.The platform is an open space for anyone, who is an air and space enthusiast and wants to spend idle computing time or even skill for a sub-project on platform. Applications for sub-project are welcome!

Project MotivationIn particular in aerospace sciences simulation and optimization are major tasks to minimize structure weight, to maximize thrust or to determine system reliability.These numerical tasks need a high level of resources and hardware-support and Constellation platform offers distributed computing power to projects from professionals to students to solve their problems in an adequate time without the need to maintain their own super-computer or cluster system.In this way it is possible to get access to the needed computing power by a wider range of researchers who wouldn't have been able to create those resources because of financial, administration and operation or because of bureaucracy reasons.Constellation's goal is to expedite fundamental and applicable research and bonding researchers with citizen scientists and public.

Project ImplementationAs a DGLR (german aerospace society)[2] academic student team at the University of Stuttgart, Germany, we set up a distributed super-computer that offers computing power by idle single desktop computers of volunteers connected via the internet. This accumulated computing power is provided to aerospace research purposes, that universities and private research institutions and projects normally can't afford. With this approach we demonstrate the capability of synergetic co-operation of scientists and scientific enthused ordinary persons.In spite of classic super-computers, that have high acquisition-, maintenance- and operation-costs and aren't state of the art after just a few years, our Constellation computer is a worldwide distributed, continuously and dynamically evolving system with high heterogeneity to which anyone can participate in who runs a computer with Window, Linux or OSX operating system. This high-performance is used to solve aerospace related tasks, such as trajectory optimization (application: TrackJack).Constellation is in its late closed test-phase and will go public in the near future. Even during test-phase 550 participating machines generated approx. 120*10^9 FLoating point OPerations per Second (120 GFLOPS). When went public it can be assumed that the amount of participating machines will increase up to 5000 with about 10 TFLOPS in one Year (reference data from RNA-World, another Rechenkraft.net e.V. project [3]).The Constellation computer represents a remarkable performance that is by far superior to high-performance clusters and it increases daily. In this way hundreds of trajectories can be simulated and optimized depending in the task's complexity per day. Participating users can follow their contribution in a clearly arranged webinterface. Finished results automatically will be send back and can be archived by the sub-projects scientists.Operation and maintenance of this system requires expenditure of time during user support, who post a variety of questions on our online-forum, so that besides the scientific aspect the project has a very interesting education component („science & society“).

Applications and PerspectivesThere are three applications as sub-projects on the Constellation platform.

o TrackJack is an ascend trajectory simulation and optimizer for space-launcher systems and space-crafts. TrackJack's origin is part of a diploma thesis in Aerospace Engineering at the University of Applied Sciences Bremen.TrackJack app is processing test-workunits and will optimize the maximum altitude of HyEnD hybrid-engine sounding rocket [4]. (Figure 2&3)

o Extreme Machine's idea is not to create a huge, multimillion ton exploration rover machine, rather an extremely optimized device, under the realms of very classical mechanics, to perform specific tasks. Extreme Machine simulates the rover dynamics. (Figure 4)

o On The Moon aims to simulate the process occurring at moon near the surface. It is a large scale simulation aiming to create a total model of the moon system, and to compare it with data collected from the Google Lunar X-PRIZE (GLXP)[5] mission.

Abstract: VOLPEX -Parallel Execution in Volunteer environmentVolpex[9-11] has the goal of enabling robust execution of communicating parallel applications on volunteer PCs. The project developed the fundamental concept of autonomous redundant processes: a logical process may have multiple distributed instances that are unaware of each other, process instances can be replicated, checkpointed or re-created independently, and individual process instances can fail without application failure. The overall application progress is dictated by the state of the fastest active instance of each process. A single replication framework can potentially provide protection from process/node and network failures, varying processing speeds, and Byzantine failures caused by software and hardware errors and malicious hosts. Two prototype programming frameworks were successfully developed and validated in this project:

o Volpex Dataspace API: An abstract ``dataspace’’ with asynchronous anonymous Put/Get communication operations from autonomous redundant processes.

o Volpex MPI: A highly failure resistant MPI implementation based on sender based logging.

A Volpex prototype has been validated and evaluated with NAS and other benchmarks, a Map-Reduce framework, and a Replica Exchange Molecular Dynamics code, and interfaces with BOINC.

ResultsEven though a lot of aerospace projects sounds like science-fiction we lay emphasis on applicable and fundamental work that will influence current research. So the participants will directly be involved and success in research can be feed back to the community to gratify them that leads to a good atmosphere and long-time participation. Constellation as a platform will leave it to the sub-projects to openly publish parts or the complete results.

References[1] BOINC – Berkeley Open Interface for Network Computing - http://boinc.berkeley.edu[2] DGLR – Deutsche Gesellschaft für Luft- und Raumfahrt - http://stuttgart.dglr.de[3] RNA World – http://rnaworld.de/rnaworld | Rechenkraft.net e.V. - http.//.rechenkraft.net[4] HyEnD – Hybrid Engine Development - http://www.hybrid-engine-development.de[5] Google Lunar X-Prize - http://glxp.org[6] MAGE – The Marburg Ad-hoc Grid Environment - http://mage.uni-marburg.de[7] OpenFOAM – Open Field Operation and Manipulation - http://www.openfoam.com[8] Ansys – CFX, Fluent - http://ansys.com[9] VOLPEX - Parallel Execution in Volunteer environment - http://volpex.cs.uh.edu/VCP[10] 'A Robust Communication Framework for Parallel Execution on Volunteer PC Grids', by Eshwar Rohit, Hien Nguyen, Nagarajan Kanna, Jaspal Subhlok, Edgar Gabriel, Qian Wang, Margaret S. Cheung, and David Anderson, The 11th IEEE/ACM International Symposium on Cluster, Cloud, and Grid Computing , Newport Beach, CA, May 2011.[11] VolpexMPI: an MPI Library for Execution of Parallel Applications on Volatile Nodes, by T. LeBlanc, R. Anand, E. Gabriel, and J. Subhlok. In Proc. The 16th EuroPVM/MPI 2009 Conference, Espoo, Finland, 2009. Published as Springer-Verlag LNCS,5759.

We want to use parallel execution for CFD analysis.We want to use OpenFOAM[7] because it's used in the scientific and academic community, it offers a wide range of solvers and the main reason is that its GNU General Public License allows the use in a distributed environment. In comparison Ansys' powerful CFX and FLUENT[8] programmes use a proprietary software licence that makes it difficult to use outside the given licence agreement.For Constellation a CFD application is intended to be used in aerodynamics and stability simulation, in engine combustion analysis for micro- and jet engines, in ramjet, scramjet and pulse detonation engine and rocket engines and motors and many more. We want to open up this important field of parallel execution for distributed computing.

Currently the BOINC infrastructure Constellation uses supports independent workunits. The above applications respect this limit by using splitted tasks as workunits that doesn't rely on other workunits' result.But there are aerospace tasks, like in computational fluid dynamics (CFD), that are only feasible when the complete tasks is solved in parallel to finish in a decent time. Therefore Constellation is working to extend the system to combine the advantages of distributed computing system with that of parallel clusters to create a virtual cluster where nodes are connected via the internet.We work together with Volpex Group at University of Houston to bring parallel execution to Constellation and to other BOINC-projects, and we examine MAGE[6] of University of Marburg as an additional candidate.

Andreas Hornig(1,2), Lars Bausch, Sayandeep Khan, Maximilian Palm, Uwe Beckert, Sebastian de Cillia, Matthias Fritzsche, S. Müller1. Constellation, König-Karl-Str. 27, 70372 Stuttgart, Germany2. corresponding author: [email protected]

http://aerospaceresearch.net

Figure 2 Figure 3 Figure 4

Figure 5: CFD analysis with FLUENT

Figure 6: possible splitted mesh processing via internet-connected client-nodes

Figure 7: VOLPEX

Figure 1: Constellation website

"Space is big. Really big. You just won't believe how vastly, hugely, mind- bogglingly big it is. I mean, you may think it's a long way down the road to the chemist's, but that's just peanuts to space."- The Hitchhiker's Guide to the Galaxy by Douglas Adams